JP2019521439A - Statistical chest model generation and personalization - Google Patents

Abstract

An image processing system having an input interface IN for receiving a plurality of acquired input images to be inspected. The system further comprises a substance type analyzer MTA configured to generate substance type readings at corresponding positions in the input image IM (CH). A statistical module SM of the system is configured to determine an estimate of a probability distribution of substance types for the corresponding location based on the readings.

Description

  The present invention relates to an image processing system, an image processing method, a computer program element and a computer readable medium.

  Biomechanical simulation of the chest can be used to simulate the chest in various patient postures, such as prone, supine or upright. Recently, biomechanical models have also been used in surgical simulations. Such applications require a precise chest geometric model. The model is usually generated from MR imaging and is a costly and time-consuming procedure. Given the prevalence of breast cancer and the fact that MRI is not a standard imaging modality, only a fraction of affected patients can benefit from preoperative simulation as decision support.

  Therefore, there may be a need for a system or method to at least partially solve the above drawbacks.

  The object of the invention is solved by the subject matter of the independent claims, and further embodiments are incorporated in the dependent claims. It should be noted that the aspects of the invention described below apply equally to image processing methods, computer program elements and computer readable media.

According to a first aspect of the invention,
An input interface for receiving a plurality of acquired input images to be inspected;
A substance type analyzer configured to generate substance type readings at corresponding positions in the input image;
A statistical module configured to determine an estimate of a probability distribution of material types for the corresponding location based on the readings;
An image processing system is provided.

  According to an embodiment, the system is further configured to correlate the estimate of the probability distribution with the metadata of the examined object to obtain an estimate of a parameterized probability distribution for the corresponding position, It has a correlator.

  According to one embodiment, the system comprises spatial counterparts configured to establish the corresponding position based on coordinates in the respective coordinate system of the image, the respective coordinate systems having a common geometry It is derived from the model.

  According to an embodiment, each said coordinate system is configured to reflect one or more symmetries in said examination object.

  According to one embodiment, the system is configured to generate an estimate of substance type for a given position in a given image of an object, based on the estimation of the parameterized probability distribution. It has a substance type estimator.

  According to one embodiment, the input image is acquired in advance by an imaging device capable of soft tissue identification.

  According to one embodiment, the test object comprises the chest of various target mammals (especially human females).

According to a further aspect,
Receiving a plurality of acquired input images to be examined;
Generating material type readings at corresponding locations in the input image;
Determining an estimate of a probability distribution of material types for the corresponding location based on the readings;
An image processing method is provided.

According to one embodiment, the method comprises
Correlating the estimate of the probability distribution with the metadata of the examined object to obtain an estimate of a parameterized probability distribution for the corresponding location.

  According to one embodiment, the corresponding position is established based on the coordinates in the respective coordinate system for the image, and the respective coordinate system is derived from a common geometric model.

According to one embodiment, the method comprises
Generating an estimate of the substance type for a given position in a given image of the object based on the parameterized probability distribution estimate.

  According to a further aspect, there is provided a computer program element configured to perform the method steps described above when being executed by a processing unit.

  According to a further aspect, a computer readable medium is provided, wherein the program element is stored.

  The proposed method and system make it possible to take into account the relatively large (in shape and composition) variability of the women's chest. The proposed method and system do not use a simple general empirical assumption about the distribution of internal tissue, but a virtual chest model that can be truly personalized / fit for any given patient It is possible to generate in a structured manner. Such simple heuristics are avoided, leading to more realistic models and thus more realistic biomechanical simulations.

  The proposed method and system make it possible to generate a distribution of tissue types as can be observed in real patients. The generation of the distribution of the tissue type is not per-place or "random", but is incorporated into a general statistical model for objects (eg chest) trained from the test subject ("cohort") Derived from scientific knowledge.

  In summary, the proposed approach is mainly applied to the chest images assumed here, statistical distribution of chest tissue in the population using high resolution modalities like MRI and CT Including training the model. By mapping each chest to a standard chest model, it is possible to generate a model of the spatial distribution of tissue. The tissue distribution model is derived from, for example, BMI, meta-information from the same population such as age, height, menopausal status, and also low-cost modalities (eg 3D scan, photography, mammography, tomosynthesis) Correlated with information. By using such information of a particular patient rather than part of a training cohort, statistical models can be later personalized into personalized chest models with appropriate size and tissue distribution. . Such personalized models allow for patient-specific biomechanical modeling of, for example, the chest, without the need for expensive tissue-identifying modalities, such as MRI, at the personalization stage.

  Embodiments of the invention are described below with reference to the drawings (not necessarily drawn to scale).

FIG. 1 shows a block diagram of an image processing system. Fig. 2 shows a flow diagram of a method of image processing.

  Referring initially to FIG. 1, a schematic block diagram of the image processing system proposed herein is shown.

  Broadly, the image processing system IPS is configured to generate a generalized statistical model of the tissue based on the input image of the tissue of interest, which model is then for the given patient PAT It may be personalized arbitrarily. In one but not necessarily all embodiments, the tissue is the chest of a human female, but other tissue may also be envisaged here. Also, applications of the proposed image processing system to non-medical situations may also be envisaged.

  A common statistical chest model is basically a spatially resolved family or collection of probability distributions for tissue types. These distributions can be personalized based on metadata such as the age, weight, etc. of a given patient PAT. General statistical models or their personalized versions can be used as input models for biomechanical simulations. Because different tissue types respond in different ways to mechanical loading, the availability of an accurate estimate of the spatial distribution of tissue types allows realistic biomechanical simulations.

  The elements or modules of the image processing system IPS may be implemented as software routines programmed to work together in a single piece of software. The system IPS may be implemented on a general purpose computing unit PU, such as a workstation associated with an imager IM, or on a server computer associated with a group of imagers. Alternatively, the components of the image processing system IPS may be configured in a distributed architecture and connected to a suitable wired or wireless communication network. Alternatively, some or all of the components of the image processing system may be configured in hardware, such as a properly programmed FPGA (field-programmable-gate-array) or as a wired IC chip .

Referring now in more detail to FIG. 1, at the input port IN of the image processing system, input images IM _j (CH) of a cohort CH of patients are received, although not necessarily simultaneously. The image has been previously acquired by one or more imagers IA. Preferably, the input image IMCH is obtained by an imager IA capable of soft tissue identification, such as a magnetic resonance imager or a phase contrast imager.

  The input image IMCH from the cohort CH of patients may be obtained, for example, from the image storage system PACS in the hospital information system HIS or the like. In another example, historical soft tissue imaging data acquired in the past from different patients may be utilized. As described in more detail below, the historical input image acts as a training corpus for comparing statistical chest models. Preferably, but not necessarily, the image is of the DICOM type. DICOM images have their respective patient metadata encoded in the header file of each image file. As described in more detail below, the metadata may be used to personalize a generic model for a particular given patient. Patients in Cohort CH can now be viewed as test subjects representing the same subject or tissue class, which in this example is that of the human chest for which a generalized statistical model is to be constructed.

As shown, the input image is used to train a statistical distribution model of chest tissue in a population of which the cohort is a sample. The statistical distribution model generated here can be expected to be more accurate as the sample is larger. Picking up cohorts randomly from women in hundreds or thousands of images results in good quality results. In one embodiment, about 100 images are used. The following processing of the input image IM (CH) _j is not necessarily a one-time operation, but can still be envisaged if a sufficient number of training images are available from the start. However, often starting with a particular initial set of input images IM (CH), then the model is refined and the process described below is repeated to take into account newly available input images good.

  The system IPS comprises a substance type analyzer MTA. The substance type analyzer includes a substance type specifying element MTC and a spatial correspondence element SCC.

  The spatial counterpart element SCC of the system IPS operates to establish each set of corresponding anatomical locations in the input cohort image IM (CH) from which the material type analyzer collects material type readings. In other words, the analyzer determines which type of material (especially which type of tissue) the image position represents.

  Referring first in detail to the spatial correspondence element SCC, in one embodiment, the world coordinates of the position in the input image are transformed into coordinates expressed in the respective coordinate systems corresponding to the symmetry of the tissue considered By doing this, a spatial or anatomical correspondence is established. For example, for the chest of a human female, D. Kutra et al. “Anatomically oriented breast model for MRI” (Proc. SPIE 9415, Medical Imaging 2015, Image-Guided Procedures Robotic Interventions, and Modeling, pp. 941521 (2015 3) It has been shown that a semi-elliptical coordinate system as proposed in Mon. 18) derives a preferred result, which can be considered here in the preferred embodiment. If other tissues are of interest, another coordinate system with a different symmetry may be called instead. Each coordinate system is derived by fitting the same shape model (geometric primitive) to each image in each fitting operation (described in more detail below). These respective coordinate systems are thus adapted to the respective geometry / symmetry encoded in the respective input image IM (CH). Representing the image position in each image, in terms of the coordinates of the coordinate system adapted to each, establishes a space-tissue correspondence in the images. In particular, the spatial correspondence element SCC makes it possible to find the spatial coordinates of the same anatomical feature in the input image. For example, the respective positions of the papilla in each of the images can be identified, as are any other anatomical positions. The same (and hence corresponding) features or anatomical positions may be expected to have the same coordinates in the respective coordinate systems of the images of the input image IMCH. For example, in a semi-elliptic coordinate system as proposed by Kurta et al. (Described above), given a coordinate (u, v, r), where u is a polar component, v is an azimuth component and r is a radial component. It can be expected that the position of the pelvis shows the same anatomical features in the chest of the cohort. However, it is noted that establishing an anatomic correspondence in the input image with an anatomically symmetric coordinate system is a preferred but merely one possible example among others. It should be. For example, instead, the user may define the location considered to represent the anatomical correspondence by mouse click on the image, or by any other designating tool.

  Continuing to refer to the behavior of the spatial counterpart, which, as noted, fits the geometry primitives (such as semi-elliptic) separately to each of the input images IM (CH) to obtain the respective geometry Including deriving a dynamically adapted image coordinate system. The respectively fitted geometric primitives approximate the shape of the chest encoded in each input image. The fitting operation establishes an anatomic patient-to-patient correspondence of the various patient images in the cohort by matching the geometric features of the geometric primitives to the image structure representing the appropriate anatomical part. For example, the tip of the semi-elliptic model may be fitted to the papilla of the image, and the wide part at the rear of the semi-elliptic model may be fitted to the image structure representing the pectoral muscle. In this way, different parts of the primitive geometric model are fixed at different prominent anatomical positions in the input image. Anatomic correspondence can be established as the same primitive geometry is utilized for fitting. In order to realize more suitable fitting, the model may be deformed by free-form deformation or the like. In other words, the fitting operation generally involves a flexible transformation.

を単に利用することである。代替としては、それぞれの幾何プリミティブをフィッティングするときにそれぞれの位置において経験される堅固な又は柔軟な変換の量に基づいて、オフセット

が座標に適用されても良い。 The world coordinates of the position of pixels or voxels representing different tissue types in each input image IM (CH) are set by the space correspondence element SCC in the case of an embodiment in which the primitives are (semi) elliptical (three sets ( Converted to geometrically unique parameters of geometric primitives, such as u, v, r). These parameters define the coordinates in the respectively adapted coordinate system based on the common geometry of the geometry primitives as used in the fitting operation. This approach makes it possible to use the generated correspondence between patients represented in the cohort image IM (CH). World coordinates include in-image coordinates (such as a voxel index), or coordinates in an appropriate length unit (e.g., mm) relative to the reference frame of the imaging system used to obtain the image. When the image format is DICOM, these types of world coordinates may be converted to each other.
As mentioned above, one way to establish correspondences between positions in an image is to use identical coordinates in each image, in each adapted coordinate system

It is simply to use Alternatively, offsets based on the amount of rigid or flexible transformation experienced at each position when fitting each geometric primitive

May be applied to the coordinates.

  The fitting operation may be implemented by any known optimization scheme. For example, the objective function may be set as the sum of the squares of the deviations of the chest image from the primitives. At this time, the geometric parameters of the primitives are solved to minimize the objective function. Other objective function forms may also be envisaged.

  Preferably, although the input image IM (CH) and the geometry primitives are 3D, application of the proposed method to 2D images by fitting of 2D models can also be considered in alternative embodiments.

  The input cohort image IM (CH) is processed by the substance type specification element MTC. This is also known automated, which makes it possible to establish substance types at each position of the input image, for example tissue types such as adipose tissue, water, fibroglandular tissue, vascular tissue, muscle tissue etc. It may be implemented by a tissue analysis / classification algorithm. Tissue analysis algorithms match image values (eg, gray values) with known tissue characteristics. For example, in particular MRI image values may be matched to the respective tissue class.

It should be understood that the two tasks of substance characterization and establishment of spatial correspondence may be performed in any order. That is, the image is first processed by the material type specific element MTC and then by the spatial counterpart SCC, or vice versa. Regardless of the processing order, the (intermediate) output of the substance type analyzer is formed from the substance type reading m _j for each set of space-tissue corresponding positions. In other words, for each set j of corresponding positions, there are N material type readings (m ₁ ,... _{M N} ) ^j , where N is the number of images in the input image IM (CH). The outputs of the spatially resolved material type readings may be stored in a matrix structure with indices j... J, n.

  The spatially resolved material type readings are processed by the statistical module SM. The statistical module calculates respective probability distribution (or density) estimates at given corresponding locations. The operation is then repeated for each set j of (other) corresponding positions. In other words, for each of the N sets of different corresponding positions, respective (generally different) probability distributions are obtained for different tissue types. Furthermore, in other words, each of the tissue type probability distributions is "fixed" to a respective set of corresponding positions expressed in intrinsic coordinates in a coordinate system respectively adapted for the image.

  The set or family of probability distributions makes it possible to calculate the probability (e.g. fat) that is there at a given position. The respective probability distributions (or probability densities) at various locations may be estimated in other known ways, for example by forming the ratio of the respective tissue type readings. For example, for a given set of corresponding positions, the number of occurrences of fat may be examined and divided by the total number of images considered. Other statistical methods may be used. For example, as an alternative, the local probability may take into account the age of the patient under consideration, BMI etc. In calculating the tissue probability, only similar objects (with respect to metadata) may be taken into account, or the contribution of data from the test cohort may be weighted by the similarity to the patient under consideration.

  Alternatively, the respective coordinates in the different adapted coordinate systems for each image may, in a second transformation, be converted to a common coordinate system (spherical or otherwise shaped) by means of further transformation stage elements (not shown). And so on). The statistics module SM then operates to analyze the spatial pattern of tissue types in the common coordinate system, for example by using a method of spatial statistics. A second transformation to a common coordinate frame allows normalization of the tissue distribution, which allows optimal visualization of the distribution.

  The set of these localized probability distributions provided by the statistical model SM forms a generalized statistical model of the chest. The generalized statistical model can be refined by the correlator CORR, which correlates these distributions to patient metadata in the patient cohort CH. Correlator CORR aims to establish a functional relationship on how these probability distributions change, as well as patient metadata of patients in the cohort, such as age, weight, BMI index etc. . Any known machine learning or curve fitting algorithm, response surface method may be used to implement the correlator CORR stage. The functional relationships calculated by the correlator CORR may be explicit (as in functional representations) or implicit as in machine learning schemes such as neural networks. The functional relationships sought may be encoded in the weights assigned to the nodes of the neural network in the various layers.

  The outputs of the correlators CORR form a parameterized family of spatially localized probability distributions obtained previously. Each parameter α represents a combination of different metadata, and these parameterized probability distributions Fα (especially the density fα) are then output at the output port OUT and are not from the cohort CH (new) patient PAT It can be used to personalize a given image IM (PAT). The parameterized probability distribution generated in the correlator CORR stage can also be understood as a conditional probability distribution. For example, the probability for fat at a given location is then conditioned on the fact that the patient has a particular age and / or BMI etc.

  The personalization operation is realized by the substance type estimator MTE. The substance type estimator MTE receives as input a new image IM (PAT) of a new patient PAT and provides an estimate of the substance type for a given position in the new image IM (PAT). Preferably, as can be expected, the new image IM (PAT) is one previously acquired by the “cheaper” imaging modality IA ′. Examples of such "cheap" imaging modalities include tomosynthesis, x-rays and the like. In another embodiment, the new image IM (PAT) may be as simple as a 3D surface scan obtained by an optical scan. In other words, the input image IM (PAT) for the substance type estimator MTE is generally of lower resolution than the cohort image IM (CH) and does not encode or generalize the tissue type It encodes only to a lower degree than the cohort image IM (CH) used for training of statistical models.

  More specifically, the new non-tissue-type-discriminating input image IM (PAT) is generated by the substance-type estimator MTE in the same manner as described above, except for tissue type analysis not possible here. It is processed. Specifically, as described above, geometric primitives (eg, semi-elliptic) are only fitted to geometric shapes, such as those encoded in the low cost image IM (PAT). Fitting of the primitives allows the identification of the position (u, v, r) using the adapted anatomical coordinate system as described above. Patient PAT metadata is obtained from the DICOM header file or others (eg by querying the patient PAT itself), which is then from the parameterized family of probability distributions provided at the output port OUT , Used to calculate a family that matches given metadata of the patient PAT. For each given position expressed in intrinsic coordinates (e.g. u, v, r), an estimate of the tissue type m (u, v, r) likely to be found at each position is created it can. The estimate is obtained from the metadata matched family of probability distributions by using one associated with the given position. For example, the substance type estimator MTE may assign the tissue type with the highest probability to the probability distribution for the position under consideration. In one embodiment, additionally, the similarity of the patient's metadata to the training data's metadata is taken into account to adjust the highest probability (eg, by weighting). From the above, the substance type estimator MTE is used in triple operations: i) finding families of probability distributions by matching against metadata, ii) matching against positions from families of probability distributions, and iii) probability distributions Perform the actual estimation. The personalized model may then be processed into a biomechanical simulation package, or may be stored in memory or otherwise processed.

  In summary, a cohort input image derived from an imaging modality capable of tissue identification is used to learn tissue identification, and the relevant knowledge of internal tissue is provided as a genetic statistical chest model. This has the advantage that the subsequently processed non-cohort image can be made to have only shape contrast and no (or not much) tissue type contrast. The non-cohort image is then used simply to learn the shape of the personalized model and then augmented with anatomical knowledge of tissue type identification such as a generalized statistical model.

  Reference is now made to FIG. 2, which shows a flow diagram of an image processing method which is a potential operation of the image processing system IPS. However, the following flow diagram can be understood on its own and is not necessarily bound by the architecture of the image processing system IPS as shown in FIG.

  In step S210, a plurality of acquired input cohort images to be examined are received. The test object is of the same class, such as the chest or other anatomical structure of (a cohort of) multiple patients. Preferably, the input cohort image is acquired by an imaging modality capable of tissue identification.

  In step S220, material type readings are generated at corresponding locations of the input cohort image.

  Step S220 includes automatically establishing the corresponding position of the input cohort image by using each coordinate system (S220a). These are derived by separately fitting a common shape model to each input cohort image. These coordinate systems are preferably adapted to one or more symmetries of image structures representing the tissue of interest recorded in the input cohort image. Preferably, the image is a chest image and the coordinate system used is semi-elliptic.

  The step of generating a substance type reading S220 further comprises identifying the substance type based on the image values in the cohort image (S220b).

  Steps S220a and S220b may be performed in any order.

  Then, in step S230, based on the readings for the corresponding position, the probability distribution of each of the substance types is determined. In this way, a dedicated probability distribution estimate is obtained for each set of corresponding positions.

  Then, in step S240, the probability distributions determined for each set of corresponding positions are correlated with the metadata to be examined. The correlating step leads to a parameterized probability estimate, each parameter representing different categories of metadata (e.g. weight, age, BMI etc). The set of parameterized probability distribution estimates for each set of corresponding positions form a generalized statistical model trained from the test object.

  The generalized statistical model may be personalized for a particular given image of a particular subject in the same class as the examination subject, but the particular subject is used to train the generalized statistical model It is not drawn from the subject of inspection. More specifically, the personalization operation is carried out in step S250, wherein the step is performed in the non-cohort image to be examined based on the parameterized probability distribution obtained in the correlation step in step S230. A substance type estimate for the given position is obtained.

  More specifically, step S250 comprises fitting the parameterized set of probability distributions to the non-tested object using the given non-tested metadata.

  The set of probability distributions so adapted is then used to estimate tissue type for non-cohort images. For example, the tissue type with the highest probability is assigned to a given location. Specifically, to match a position in the non-cohort image to the corresponding probability distribution (from the set) for that position, any geometry primitives used above in step S220 are fitted to the non-cohort image Ru. Then, a coordinate system based on the fitted model is utilized to define i) a given position in the non-cohort image, and ii) a fitted probability distribution from the set for that position. Step S240 is then repeated for each position to obtain a personalized model for a given object. In other words, the personalized model has a fitted shape and an internal tissue distribution, like a fitted probability distribution.

  At any step, the personalized model may be sent to biomechanical simulation tools or processed.

  In the foregoing, the cohort image includes tissue contrast but does not include non-cohort images or only to a lesser extent. In particular, non-cohort images are simply sufficient to include shape information. In one example, non-cohort images may be as simple as surface scans or x-ray images, but cohort images are obtained from tissue imaging modalities such as MRI etc.

  Preferably, in the above, the input cohort image and the new specific non-cohort image are acquired for a patient who takes the same attitude against gravity. For example, the cohort input image and the new image are acquired in a prone position.

  The end result of the method is a personalized anatomical software / virtual chest model, for example with the external shape of the chest in the prone position and the distribution of internal structures such as fibroglandular tissue, fat etc. . The personalized model may be utilized in biomechanical software simulation packages such as finite element packages such as niftysym, FeBio etc.

  In another embodiment of the present invention there is provided a computer program or computer program element characterized in that it is arranged to carry out the method steps of the method according to one of the above embodiments on a suitable system.

  Thus, the computer program element may be stored on a computer unit which may be part of an embodiment of the invention. The computer unit may be configured to perform or direct the execution of the steps of the method described above. Furthermore, the computer unit may be configured to operate the components of the device described above. The computer unit may be configured to operate automatically and / or may be configured to execute user commands. The computer program may be loaded into the working memory of the data processor. Thus, a data processor may be provided to carry out the method of the invention.

  The present embodiment of the present invention covers both a computer program which utilizes the present invention from the beginning and a computer program which, upon update, switches an existing program to a program which utilizes the present invention.

  Furthermore, the computer program element may be able to provide all the steps necessary to carry out the procedure of the embodiment of the invention described above.

  According to a further embodiment of the present invention, there is provided a computer readable medium such as a CD-ROM, wherein the computer readable medium is a storage of computer program elements as described in the preceding sections. is there.

  The computer program may be stored / distributed on a suitable medium, such as an optical storage medium or solid medium supplied with other hardware or as part of other hardware, but the Internet or other wired or wired It may be distributed in other forms, such as via a wireless communication system.

  However, the computer program may be provided through a network such as the World Wide Web, and may be downloaded from such a network to the working memory of the data processor. According to a further embodiment of the present invention there is provided a medium for making a computer program element available for download, said computer program element according to one of the above described embodiments of the present invention. It is configured to carry out the method.

  It should be noted that the embodiments of the present invention have been described in relation to various subjects. In particular, some embodiments are described with respect to method type claims, and other embodiments are described with respect to apparatus type claims. However, those skilled in the art will appreciate that, from the foregoing and following descriptions, any combination of features relating to various subjects, in addition to any combination of features belonging to one type of the invention, unless otherwise stated. It will be understood that this is considered to be the disclosure of this specification. However, any feature may be combined to provide a synergetic effect beyond just the sum of these features.

  While the present invention has been illustrated and described in the drawings and the foregoing description, such description and description are to be considered illustrative or exemplary and not restrictive. The invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention by reading the drawings, the description and the dependent claims.

  In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may perform the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures can not be used to advantage. Any reference signs in the claims should not be construed as limiting the scope of the claims.

Claims (13)

An input interface for receiving a plurality of acquired input images to be inspected;
A substance type analyzer configured to generate substance type readings at corresponding positions in the input image;
A statistical module configured to determine an estimate of a probability distribution of material types for the corresponding location based on the readings;
An image processing system having   The system according to claim 1, comprising a correlator configured to correlate the estimate of the probability distribution with the metadata of the examined object to obtain an estimate of a parameterized probability distribution for the corresponding position. .   3. A space-corresponding element configured to establish the corresponding position based on coordinates in a respective coordinate system for the image, the respective coordinate system being derived from a common geometric model. The system described in.   The system of claim 3, wherein each of the coordinate systems is configured to reflect one or more symmetries in the examination subject.   A material type estimator configured to generate a material type estimate for a given position in a given image of an object based on the parameterized probability distribution estimate. The system according to any one of 4.   The system according to any one of claims 1 to 5, wherein the input image is acquired in advance by an imaging device capable of soft tissue identification.   7. The test subject according to any one of claims 1 to 6, wherein the test subject includes the breasts of various target mammals. Receiving a plurality of acquired input images to be examined;
Generating material type readings at corresponding locations in the input image;
Determining an estimate of a probability distribution of material types for the corresponding location based on the readings;
And an image processing method. 9. The method of claim 8, comprising correlating the probability distribution estimate with the examined metadata to obtain an estimate of a parameterized probability distribution for the corresponding location.   The method according to claim 8 or 9, wherein the corresponding position is established based on coordinates in the respective coordinate system for the image, and the respective coordinate system is derived from a common geometric model. 11. A method according to any one of claims 8 to 10, comprising generating an estimate of material type for a given position in a given image of an object based on the estimation of said parameterized probability distribution. the method of.   A computer program element configured to perform the method according to any one of claims 8 to 11 when executed by a processing unit.   A computer readable medium, wherein the program element according to claim 12 is stored.

Images